Nombre: JOSÉ MARQUES DE OLIVEIRA JÚNIOR
Tipo: MSc dissertation
Fecha de publicación: 04/04/2022
Supervisor:
Nombre |
Papel |
|---|---|
| CELSO JOSE MUNARO | Advisor * |
| PATRICK MARQUES CIARELLI | Co-advisor * |
Junta de examinadores:
| Nombre |
Papel |
|---|---|
| CELSO JOSE MUNARO | Advisor * |
| LUIZ AFFONSO HENDERSON GUEDES DE OLIVEIRA | External Examiner * |
| PATRICK MARQUES CIARELLI | Co advisor * |
| RICARDO EMANUEL VAZ VARGAS | External Examiner * |
Sumario: Produced water, on offshore platforms, is one of the effluents recovered from wells together with oil and natural gas, being the main waste generated in this process. The Total Oil and Grease (TOG) is considered one of the main parameters for controlling the disposal of produced water at sea, with daily and monthly limits defined by current legislation. The TOG measurement used as
a reference by IBAMA is done by the gravimetric method, with water samples collected daily and sent to an accredited laboratory, which provides the results with a lag of a few days from the sampling date. The need for corrective actions in case of values above the limit has motivated the use of alternative methods that generate estimates more frequently. In this work, data-based
models are created to obtain TOG estimates. Produced water treatment process variables, chemical information and daily production data from an offshore platform were collected, treated and used to train, validate and test these models. In addition, hyperparameter optimization and feature selection techniques were applied. The results obtained showed that the models based on recurrent neural networks (LSTM and CNN+LSTM) achieved superior performances compared to existing online monitoring systems.
